src/local_transcriber/backends/onnx_asr.py

@@ -8,7 +8,7 @@ from dataclasses import dataclass
 from pathlib import Path
 from typing import Any
 
-from local_transcriber.types import Segment, TranscribeResult
+from local_transcriber.types import UNKNOWN_LANGUAGE, Segment, TranscribeResult
 
 
 @dataclass(frozen=True)
@@ -216,7 +216,9 @@ class OnnxAsrBackend:
 
         _notify(on_status, "Транскрибирую (onnx-asr)...")
         segments: list[Segment] = []
-        result_language = language or _model_language(self._model_spec) or "unknown"
+        result_language = (
+            language or _model_language(self._model_spec) or UNKNOWN_LANGUAGE
+        )
 
         for vad_seg in model.recognize(
             audio_array, sample_rate=16000, language=language

src/local_transcriber/backends/openvino.py

@@ -12,7 +12,7 @@ from typing import Any
 from huggingface_hub import snapshot_download
 from huggingface_hub.errors import LocalEntryNotFoundError
 
-from local_transcriber.types import Segment, TranscribeResult
+from local_transcriber.types import UNKNOWN_LANGUAGE, Segment, TranscribeResult
 
 # (model_alias, compute_type) → HF repo
 MODEL_REPOS: dict[tuple[str, str], str] = {
@@ -159,7 +159,7 @@ class OpenVINOBackend:
                     f"Транскрибирую (OpenVINO)... [{len(segments)} сегм.]",
                 )
 
-        detected_language = language or "auto"
+        detected_language = language or UNKNOWN_LANGUAGE
         language_probability = 1.0 if language else 0.0
 
         return TranscribeResult(

src/local_transcriber/cli.py

@@ -12,6 +12,10 @@ from .config import apply_device_defaults, load_config, resolve_defaults
 from .context_menu import install_menu as install_context_menu
 from .context_menu import uninstall_menu as uninstall_context_menu
 from .formatter import (
+    LANGUAGE_DETECTED,
+    LANGUAGE_FORCED,
+    LANGUAGE_FROM_MODEL,
+    LANGUAGE_UNKNOWN,
     format_duration,
     format_timestamp,
     format_transcript,
@@ -30,6 +34,7 @@ from .transcriber import (
     _transcribe_file,
     load_model,
 )
+from .types import UNKNOWN_LANGUAGE
 from .utils import (
     build_output_path,
     detect_device,
@@ -62,12 +67,12 @@ def _format_language_mode(
 ) -> str:
     """Описывает источник языка, не выдавая профиль модели за детектор."""
     if requested_language != "auto":
-        return "forced"
+        return LANGUAGE_FORCED
     if result.language_probability > 0:
-        return "detected"
-    if result.language not in {"", "auto", "unknown"}:
-        return "из профиля модели"
-    return "не определён"
+        return LANGUAGE_DETECTED
+    if result.language not in {"", UNKNOWN_LANGUAGE}:
+        return LANGUAGE_FROM_MODEL
+    return LANGUAGE_UNKNOWN
 
 
 def _format_repetition_blocks(

src/local_transcriber/formatter.py

@@ -10,6 +10,19 @@ from .types import Segment, TranscribeResult
 _PAUSE_THRESHOLD_S = 2.0  # пауза между сегментами для разбиения на абзацы
 _MAX_PARAGRAPH_S = 60.0  # максимальная длительность абзаца
 
+# Источник языка в шапке транскрипта
+LANGUAGE_FORCED = "задан явно"
+LANGUAGE_DETECTED = "определён автоматически"
+LANGUAGE_FROM_MODEL = "из профиля модели"
+LANGUAGE_UNKNOWN = "не определён"
+
+LANGUAGE_MODES = (
+    LANGUAGE_FORCED,
+    LANGUAGE_DETECTED,
+    LANGUAGE_FROM_MODEL,
+    LANGUAGE_UNKNOWN,
+)
+
 
 @dataclass
 class _Paragraph:
@@ -80,7 +93,7 @@ def format_transcript(
     source_filename: str,
     model_name: str,
     device_info: str,
-    language_mode: str,  # detected | forced | из профиля модели | не определён
+    language_mode: str,  # см. LANGUAGE_MODES
     transcription_date: datetime | None = None,  # None -> datetime.now()
 ) -> str:
     """Собирает markdown-транскрипт: шапка с метаданными + абзацы с таймкодами."""
@@ -92,6 +105,9 @@ def format_transcript(
     lines.append("")
     lines.append(f"- **Дата транскрипции**: {date.strftime('%Y-%m-%d %H:%M:%S')}")
     lines.append(f"- **Модель**: {model_name}")
+    if language_mode == LANGUAGE_UNKNOWN:
+        lines.append(f"- **Язык**: {LANGUAGE_UNKNOWN}")
+    else:
         lines.append(f"- **Язык**: {result.language} ({language_mode})")
     lines.append(f"- **Длительность**: {format_duration(result.duration)}")
     if tail_gap(result) > TAIL_GAP_WARN_S:

src/local_transcriber/types.py

@@ -4,6 +4,9 @@ from collections.abc import Callable
 from dataclasses import dataclass
 from typing import Any
 
+# Единый признак «язык неизвестен» для всех бэкендов
+UNKNOWN_LANGUAGE = "unknown"
+
 
 @dataclass
 class Segment:
@@ -15,7 +18,7 @@ class Segment:
 @dataclass
 class TranscribeResult:
     segments: list[Segment]
-    language: str
+    language: str  # код языка или UNKNOWN_LANGUAGE, если он неизвестен
     language_probability: float
     duration: float  # seconds
     device_used: str  # "cpu" / "cuda" / "onnx" / "openvino-gpu" / "openvino-cpu"

tests/test_backend_openvino.py

@@ -11,7 +11,7 @@ from local_transcriber.backends.openvino import (
     OpenVINOBackend,
     _validate_model_dir,
 )
-from local_transcriber.types import Segment
+from local_transcriber.types import UNKNOWN_LANGUAGE, Segment
 
 
 # === _resolve_repo ===
@@ -301,7 +301,7 @@ def test_transcribe_no_language_auto():
 
     call_kwargs = mock_model.generate.call_args.kwargs
     assert "language" not in call_kwargs
-    assert result.language == "auto"
+    assert result.language == UNKNOWN_LANGUAGE
     assert result.language_probability == 0.0
 
 

tests/test_cli.py

@@ -6,7 +6,14 @@ from rich.console import Console
 from typer.testing import CliRunner
 
 from local_transcriber.cli import _format_device_info, _format_language_mode, app
+from local_transcriber.formatter import (
+    LANGUAGE_DETECTED,
+    LANGUAGE_FORCED,
+    LANGUAGE_FROM_MODEL,
+    LANGUAGE_UNKNOWN,
+)
 from local_transcriber.transcriber import Segment, TranscribeFileResult, TranscribeResult
+from local_transcriber.types import UNKNOWN_LANGUAGE
 
 runner = CliRunner()
 
@@ -45,10 +52,10 @@ def _make_tfr(result=None, model=None, actual_device="cpu", backend=None, model_
 @pytest.mark.parametrize(
     ("requested_language", "language", "probability", "expected"),
     [
-        ("ru", "ru", 1.0, "forced"),
-        ("auto", "ru", 0.95, "detected"),
-        ("auto", "ru", 0.0, "из профиля модели"),
-        ("auto", "unknown", 0.0, "не определён"),
+        ("ru", "ru", 1.0, LANGUAGE_FORCED),
+        ("auto", "ru", 0.95, LANGUAGE_DETECTED),
+        ("auto", "ru", 0.0, LANGUAGE_FROM_MODEL),
+        ("auto", UNKNOWN_LANGUAGE, 0.0, LANGUAGE_UNKNOWN),
     ],
 )
 def test_format_language_mode(

tests/test_formatter.py

@@ -2,12 +2,16 @@ from datetime import datetime
 from pathlib import Path
 
 from local_transcriber.formatter import (
+    LANGUAGE_DETECTED,
+    LANGUAGE_FORCED,
+    LANGUAGE_UNKNOWN,
     _group_segments,
     format_timestamp,
     format_transcript,
     write_transcript,
 )
 from local_transcriber.transcriber import Segment, TranscribeResult
+from local_transcriber.types import UNKNOWN_LANGUAGE
 
 
 def test_format_timestamp_minutes():
@@ -40,14 +44,14 @@ def test_format_transcript_basic():
         source_filename="meeting.mp4",
         model_name="large-v3",
         device_info="CUDA (NVIDIA GeForce RTX 3060)",
-        language_mode="detected",
+        language_mode=LANGUAGE_DETECTED,
         transcription_date=datetime(2026, 3, 17, 14, 30, 5),
     )
 
     assert "# Транскрипт: meeting.mp4" in content
     assert "**Дата транскрипции**: 2026-03-17 14:30:05" in content
     assert "**Модель**: large-v3" in content
-    assert "**Язык**: ru (detected)" in content
+    assert "**Язык**: ru (определён автоматически)" in content
     assert "**Длительность**: 02:00" in content
     assert "**Устройство**: CUDA (NVIDIA GeForce RTX 3060)" in content
     assert "---" in content
@@ -55,6 +59,27 @@ def test_format_transcript_basic():
     assert "[00:00.00 - 00:09.15] Добрый день, коллеги. Первый вопрос." in content
 
 
+def test_format_transcript_unknown_language_without_placeholder():
+    """Неизвестный язык печатается одной строкой, без служебного значения."""
+    result = TranscribeResult(
+        segments=[Segment(start=0.0, end=4.0, text=" Добрый день.")],
+        language=UNKNOWN_LANGUAGE,
+        language_probability=0.0,
+        duration=120.0,
+        device_used="openvino-cpu",
+    )
+    content = format_transcript(
+        result,
+        source_filename="meeting.mp4",
+        model_name="medium",
+        device_info="OpenVINO (CPU)",
+        language_mode=LANGUAGE_UNKNOWN,
+    )
+
+    assert "**Язык**: не определён" in content
+    assert UNKNOWN_LANGUAGE not in content
+
+
 def test_format_transcript_segment_no_leading_space():
     """Сегменты без ведущего пробела должны форматироваться корректно."""
     result = TranscribeResult(
@@ -69,7 +94,7 @@ def test_format_transcript_segment_no_leading_space():
         source_filename="f.mp3",
         model_name="tiny",
         device_info="CPU",
-        language_mode="detected",
+        language_mode=LANGUAGE_DETECTED,
         transcription_date=datetime(2026, 1, 1, 0, 0, 0),
     )
     assert "[00:00.00 - 00:02.00] Hello" in content
@@ -88,7 +113,7 @@ def test_format_transcript_empty():
         source_filename="silence.wav",
         model_name="tiny",
         device_info="CPU",
-        language_mode="detected",
+        language_mode=LANGUAGE_DETECTED,
         transcription_date=datetime(2026, 1, 1, 0, 0, 0),
     )
 
@@ -113,12 +138,12 @@ def test_format_transcript_long():
         source_filename="long.mp4",
         model_name="large-v3",
         device_info="CUDA",
-        language_mode="forced",
+        language_mode=LANGUAGE_FORCED,
         transcription_date=datetime(2026, 3, 17, 10, 0, 0),
     )
 
     assert "**Длительность**: 01:03:20" in content
-    assert "**Язык**: en (forced)" in content
+    assert "**Язык**: en (задан явно)" in content
     # Timestamps should use hours format
     assert "[00:00:00.00 - 00:00:10.50] Начало." in content
     assert "[01:01:40.00 - 01:01:50.25] Конец." in content
@@ -188,7 +213,7 @@ def test_format_transcript_tail_gap_warning():
         source_filename="tail.mp3",
         model_name="medium",
         device_info="CPU",
-        language_mode="forced",
+        language_mode=LANGUAGE_FORCED,
         transcription_date=datetime(2026, 1, 1, 0, 0, 0),
     )
 
@@ -210,7 +235,7 @@ def test_format_transcript_no_tail_gap_warning_for_small_gap():
         source_filename="ok.mp3",
         model_name="medium",
         device_info="CPU",
-        language_mode="forced",
+        language_mode=LANGUAGE_FORCED,
         transcription_date=datetime(2026, 1, 1, 0, 0, 0),
     )
 
@@ -231,7 +256,7 @@ def test_format_transcript_no_tail_gap_warning_for_exact_threshold():
         source_filename="ok.mp3",
         model_name="medium",
         device_info="CPU",
-        language_mode="forced",
+        language_mode=LANGUAGE_FORCED,
         transcription_date=datetime(2026, 1, 1, 0, 0, 0),
     )
 
@@ -257,7 +282,7 @@ def test_format_transcript_repetition_warning():
         source_filename="repeat.mp3",
         model_name="medium",
         device_info="CPU",
-        language_mode="forced",
+        language_mode=LANGUAGE_FORCED,
         transcription_date=datetime(2026, 1, 1, 0, 0, 0),
     )
 
@@ -284,7 +309,7 @@ def test_format_transcript_repetition_warning_uses_hours():
         source_filename="long-repeat.mp3",
         model_name="medium",
         device_info="CPU",
-        language_mode="forced",
+        language_mode=LANGUAGE_FORCED,
         transcription_date=datetime(2026, 1, 1, 0, 0, 0),
     )
 
@@ -305,7 +330,7 @@ def test_format_transcript_without_anomalies_has_no_warning_lines():
         source_filename="ok.mp3",
         model_name="medium",
         device_info="CPU",
-        language_mode="forced",
+        language_mode=LANGUAGE_FORCED,
         transcription_date=datetime(2026, 1, 1, 0, 0, 0),
     )
 

tests/test_onnx_asr.py

@@ -5,7 +5,7 @@ import warnings
 import pytest
 
 from local_transcriber.backends.onnx_asr import OnnxAsrBackend
-from local_transcriber.types import Segment, TranscribeResult
+from local_transcriber.types import UNKNOWN_LANGUAGE, Segment, TranscribeResult
 
 
 class FakeVadSegment:
@@ -390,7 +390,7 @@ class TestTranscribe:
         result = backend.transcribe(FakeModel(), wav_file, language=None)
 
         assert len(result.segments) == 0
-        assert result.language == "unknown"
+        assert result.language == UNKNOWN_LANGUAGE
         assert result.duration == 1.0
 
     def test_transcribe_skips_zero_length_vad_segments(self, monkeypatch, tmp_path):